Introduction Data and TFP Measurement Analysis and Results What’s behind dispersion? Conclusions Productivity dispersion and slowdown: a knowledge diffusion story? Dany Bahar (Brookings and Harvard CID) Sebastian Strauss (Brookings) KDI-Brookings Workshop January 2017 Bahar and Strauss Productivity and Knowledge KDI-Brookings Jan ’17 1 / 37
Figure: Total Factor Productivity by year (world distribution) 1.4 TFP at constant national prices (2011=1) 1.2 1 .8 .6 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 Source: Penn World Tables 9.0
Introduction Data and TFP Measurement Analysis and Results What’s behind dispersion? Conclusions Motivation Motivation and Research Question Overall slowdown in aggregate productivity (e.g. Syverson, 2016; Bailey and Montalbano, 2016; Andrews et al. 2016): Mismeasurement (see discussion in Syverson, 2016); Slowdown in supply of innovation (e.g. Gordon, 2015; Cowen, 2011); Decline in firm adoption and diffusion rates for new technologies (e.g. Comin and Hobijn, 2010; Anzoategui et al. 2015) Within industry productivity dispersion is high (e.g. Hsieh and Klenow, 2003; Syverson, 2004) This paper studies (i) dynamics of productivity dispersion; (ii) link between dispersion and productivity growth; (iii) determinants of productivity dispersion Bahar and Strauss Productivity and Knowledge KDI-Brookings Jan ’17 3 / 37
Introduction Data and TFP Measurement Analysis and Results What’s behind dispersion? Conclusions Findings Main findings Using standardized firm-level for 17M firms in ~40 countries I find robust "convergence-divergence" TFP growth dynamics: Convergence up to firms in 99th percentile of TFP: 99th percentile’s estimated 3-yr CAGR of 99th is up to ~5pp higher than for 75th percentile U-shaped convergence is stronger for developing countries, as well as for manufacturing and for IT and financial services sectors Initial levels of dispersion negatively correlates w/ growth rates a 1 s.d. increase in dispersion is a associated to ~1pp points decrease in CAGR Bahar and Strauss Productivity and Knowledge KDI-Brookings Jan ’17 4 / 37
Introduction Data and TFP Measurement Analysis and Results What’s behind dispersion? Conclusions Findings Main findings (cont) Find evidence suggesting TFP dispersion decreases are better explained for: dynamic industries (prone to competition) under regulation that favors domestic competition less so for foreign competition... knowledge intensive industries in environments more beneficial to innovation and entrepreneurship high labor mobility industries under regulations that promote talent attraction and retention less so for flexibility in labor markets ICT intensive industries in countries with higher educational attainment (weak) Bahar and Strauss Productivity and Knowledge KDI-Brookings Jan ’17 5 / 37
Introduction Data and TFP Measurement Analysis and Results What’s behind dispersion? Conclusions Findings Roadmap Data and Productivity Measurement Results Concluding Remarks Bahar and Strauss Productivity and Knowledge KDI-Brookings Jan ’17 6 / 37
Introduction Data and TFP Measurement Analysis and Results What’s behind dispersion? Conclusions Data Data Main source: Bureau Van Dijk’s Orbis dataset other papers using it are Bloom et al. 2012, Fons-Rosen et al. 2013 We complement US firms using COMPUSTAT Final database includes complete information to compute TFP for ~17M firms in 38 countries and ~500 6-digit NAICS codes, between years 2006 to 2014 Country List Bahar and Strauss Productivity and Knowledge KDI-Brookings Jan ’17 7 / 37
Introduction Data and TFP Measurement Analysis and Results What’s behind dispersion? Conclusions Data Countries in the Sample Figure: Countries in the Sample Bahar and Strauss Productivity and Knowledge KDI-Brookings Jan ’17 8 / 37
Introduction Data and TFP Measurement Analysis and Results What’s behind dispersion? Conclusions Data Estimating TFP We estimate elasticities for labor, capital and materials using 4 methodologies: Ordinary Least Squares, per country and 3-digit industries Levinsohn and Petrin (2003) following Wooldrige (2009), per country and 3-digit industries, using 1, 2 and 3 lags as instrumental variables Computing average cost shares per country, year and 3 digit industry Computing plant-specific cost shares All measures are highly correlated. We stick in this presentation to industry-level cost shares, since it provides the largest sample. Bahar and Strauss Productivity and Knowledge KDI-Brookings Jan ’17 9 / 37
Introduction Data and TFP Measurement Analysis and Results What’s behind dispersion? Conclusions Data Different TFP Measures Table: TFP Correlation Table Variables LnTFP (Cost Shares Ind) LnTFP (Cost Shares Plant) LnTFP (LP 2 lags) LnTFP (LP 3 lags) LnTFP (OLS) LnTFP (Cost Shares Ind) 1.000 LnTFP (Cost Shares Plant) 0.891 1.000 LnTFP (LP 2 lags) 0.571 0.504 1.000 LnTFP (LP 3 lags) 0.558 0.491 0.958 1.000 LnTFP (OLS) 0.608 0.529 0.745 0.739 1.000 Bahar and Strauss Productivity and Knowledge KDI-Brookings Jan ’17 10 / 37
Introduction Data and TFP Measurement Analysis and Results What’s behind dispersion? Conclusions Data Different TFP Measures Figure: TFP comparison using different methods 4 Total Factor Productivity (log) 2013 2 0 −2 LnTFP (LP 2 lags) LnTFP (LP 3 lags) LnTFP (OLS) LnTFP (Cost Shares Ind) LnTFP (Cost Shares Plant) For visualization purposes, the graph excludes severe outliers in the distribution. Bahar and Strauss Productivity and Knowledge KDI-Brookings Jan ’17 11 / 37
Introduction Data and TFP Measurement Analysis and Results What’s behind dispersion? Conclusions Descriptive Descriptive Stats Figure: TFP Distribution by Industry, 2008-2013 4 Total Factor Productivity (log) 2 0 −2 −4 s g n g e e e n o n c c i i i i r n r h t r e t c u a s s m u i u t n F r c d t a m F i / s n e f o n u & I r C u o n l l T A t C a u l I / M c s i e r i g t i A i l t U g / n i n n i M Bahar and Strauss Productivity and Knowledge KDI-Brookings Jan ’17 12 / 37
Introduction Data and TFP Measurement Analysis and Results What’s behind dispersion? Conclusions Descriptive Descriptive Stats Table: TFP Dispersion Statistics (all years) Measure Mean 25th Median 75th Std. Dev. IQ ratio 2.42 1.56 1.93 2.61 2.56 99th-Median ratio 5.95 2.91 4.14 6.25 39.88 90th-10th ratio 8.97 2.58 4.09 7.73 67.31 95th-5th ratio 29.41 3.92 7.37 17.44 302.43 Bahar and Strauss Productivity and Knowledge KDI-Brookings Jan ’17 13 / 37
Introduction Data and TFP Measurement Analysis and Results What’s behind dispersion? Conclusions Descriptive Descriptive Stats Table: TFP Dispersion Statistics, Manufacturing only (all years) Measure Mean 25th Median 75th Std. Dev. IQ ratio 1.78 1.50 1.66 1.96 0.40 99th-Median ratio 3.68 2.42 3.11 4.03 3.03 90th-10th ratio 3.61 2.32 2.88 4.11 2.33 95th-5th ratio 6.97 3.30 4.49 7.42 8.88 Bahar and Strauss Productivity and Knowledge KDI-Brookings Jan ’17 14 / 37
Introduction Data and TFP Measurement Analysis and Results What’s behind dispersion? Conclusions Descriptive Transition Matrix Table: TFP Transition Matrix PPPPPPPP 2013 1 2 3 4 5 2008 P 1 0.55 0.26 0.14 0.08 0.06 2 0.23 0.36 0.26 0.15 0.08 3 0.10 0.22 0.31 0.26 0.14 4 0.06 0.10 0.21 0.33 0.26 5 0.05 0.06 0.09 0.18 0.45 Bahar and Strauss Productivity and Knowledge KDI-Brookings Jan ’17 15 / 37
Introduction Data and TFP Measurement Analysis and Results What’s behind dispersion? Conclusions Descriptive Transition Matrix (Manufacturing Only) Table: TFP Transition Matrix (Manufacturing Only) PPPPPPPP 2013 1 2 3 4 5 2008 P 1 0.56 0.26 0.13 0.07 0.06 2 0.24 0.36 0.26 0.15 0.08 3 0.10 0.23 0.31 0.26 0.13 4 0.06 0.11 0.22 0.33 0.25 5 0.04 0.04 0.08 0.19 0.48 Bahar and Strauss Productivity and Knowledge KDI-Brookings Jan ’17 16 / 37
Introduction Data and TFP Measurement Analysis and Results What’s behind dispersion? Conclusions Growth Growth Regressions Is the slow down in productivity growth across the board? Can we identify any pattern that relates to the increasing dispersion over the years? I run a simple growth regression: TFP _ CAGR f , c , i , t → T = β × TFP f , t + η c , t + ϕ i , t + ε f , c , i , t where T − t = { 3 , 5 } . We allow for a quadratic term to explore non-linearities. Bahar and Strauss Productivity and Knowledge KDI-Brookings Jan ’17 17 / 37
Introduction Data and TFP Measurement Analysis and Results What’s behind dispersion? Conclusions Growth Results: Growth Regression Table: Growth Regression Dependent Variable: TFP Growth Rate (CAGR) 3 years 5 years (1) (2) (3) (4) lnTFP -0.2049 -0.4638 -0.1344 -0.2651 (0.040)*** (0.107)*** (0.032)*** (0.081)*** lnTFP_sq 0.0900 0.0441 (0.026)*** (0.018)** N 2185821 2185821 937384 937384 r2 0.07 0.09 0.16 0.18 All columns include country-year and product-year fixed effects. Standard errors clustered at the country and product level are presented in parenthesis. ∗ p < 0 . 10 , ∗∗ p < 0 . 05 , ∗∗∗ p < 0 . 01 Bahar and Strauss Productivity and Knowledge KDI-Brookings Jan ’17 18 / 37
Recommend
More recommend
